Sustaining Internet with Hyperbolic Mapping - Marian Boguna, Fragkiskos Papadopoulos & Dmitri Krioukov
Jun 09, 2015
Sustaining Internet with Hyperbolic Mapping
- Marian Boguna, Fragkiskos Papadopoulos & Dmitri Krioukov
29/8/13 Hyperbolic Mapping 2
Overview
• Hyperbolic Map –basic philosophy• Routing Scheme – Greedy Forwarding• Results-stretch, %shortest path, RT• Model to map current Internet on hyperbolic
Space
29/8/13 Hyperbolic Mapping 3
Hyperbolic Map-Basic Philosophy
• Coordinate system in geometric space!• Assign AS coordinates in some geometric space. • Use the space to forward information packets in the
right directions towards their destinations.
• Hyperbolic Map• Angular coordinate- as per geography• Radial coordinate – as a function of the AS degree,
making the space hyperbolic• Only information that ASs must maintain is the
coordinates of their neighbours
30/08/13 Hyperbolic Mapping 4
Greedy routing (Kleinberg)
• Greedy forwarding implements routing in the right direction: upon reading the destination address in the packet, the current packet holder forwards the packet to its neighbour closest to the destination in the space
• Uses local neighborhood information only.
Advantages- Greedy Routing
• Greedy routing in complex networks, like the real AS Internet, embedded in hyperbolic spaces, is always successful and always follows shortest paths
• Even if some links are removed, emulating topology dynamics, greedy routing finds remaining paths if they exist.
• No recomputation of node coordinates required.
Results
• Percentage of successful greedy paths 99.99% • Avg Stretch - 1.1• Routing Information at AS AS Degree • approximately the same traffic load on nodes as
shortest-path forwarding• Percentage of successful greedy paths after removal of
X% of links or nodes• X=10% 99%• X=30% 95%
The Model:Einsteinian
• The main property of hyperbolic geometry is the exponential expansion of space expansion of space
• Angular node density
is uniform, but in radial
the number of nodes
grows exponentially
as we move away from
the origin.
29/8/13 Hyperbolic Mapping 9
Einsteinian Model
By hyperbolic law of cosines,
Length(ab) is –
coshxab = coshra coshrb-sinhrasinhrbcosab;
ab – is the angle between Oa and Ob.
• rO = 0 in formula gives, Oa = ra and Ob = rb.
29/8/13 Hyperbolic Mapping 10
Methods
•
10/5/06 Lecture #12: Inter-Domain Routing 11
Methods Cont..
•
10/5/06 Lecture #12: Inter-Domain Routing 12
Methods Cont..
•
10/5/06 Lecture #12: Inter-Domain Routing 13
Internet Topology
• 23572 Ass, 58416 AS links• avg AS degree=4.92, and max degree=2778• avg clustering=0.61• hyperbolic disc radius=27• power law exponent=2.1.
30/08/13 Hyperbolic Mapping 16
Mapping AS to countries
• From CAIDA AS ranking project. • Uses 2 methods –• 1) IP based: splits the IP address space
advertised by an AS into small blocks, and then maps each block to a country.
• 2) IP and WHOIS based: reports the country where the AS headquarters are located according to the WHOIS database, for AS’s located in many countries.
30/08/13 Hyperbolic Mapping 17
Geographic Routing
• AS angular coordinates equal to their geographic coordinates, the radial coordinate is obtained by Einsteinian model according to the relationship between node degrees and radial positions.
• Then greedy forwarding in the three-dimensional hyperbolic space is performed for Routing.
30/08/13 Hyperbolic Mapping 18
Traffic & Congestion considerations
30/08/13 Hyperbolic Mapping 19
success ratio& average stretchOn removal of a given fraction of AS nodes
20
30/08/13 Hyperbolic Mapping 21
Thank You
Metropolis-Hastings
• Compute current likelihood Lc
• Select a random node• Move it to a new random angular coordinate• Compute new likelihood Ln
• If Ln > Lc, accept the move
• If not, accept it with probability Ln / Lc
• Repeat
ji
aij
aij
ijij xpxpL 1)](1[)(
Sensitivity to missing links
30/08/13 Hyperbolic Mapping 23
Mapping the real Internetusing statistical inference methods
• Measure the Internet topology properties• Map them to model parameters• Place nodes at hyperbolic coordinates (r,)
• ’s are uniformly distributed on [0,2]• Apply the Metropolis-Hastings algorithm to find ’s
maximizing the likelihood that Internet is produced by the model